Seminar: Imaging and Lithology Classification on Ernest Henry Au-Cu Deposit hyperspectral and RGB Data, 21st April, 1PM

When: Thursday 21st of April, 1PM AEDT

Where: The talk will be presented online via Zoom, RSVP here.

Speaker: Shaoqiu Zheng

Title: Imaging and Lithology Classification on Ernest Henry Au-Cu Deposit hyperspectral and RGB Data

Abstract:

Automated processing of hyperspectral data can provide an efficient, non-destructive and reproducible method of identifying the mineral in exploration drill core. However, the texture or lithology of the rock that is visible in the cores can provide additional information about the zonation and alteration present in the deposit. These are important when interpreting the structure and shape of a deposit, but cannot be interpreted from line-scan hyperspectral data. Our study considers the Ernest Henry Cu-Au deposit in Northern Queensland, Australia. Hyperspectral visible near-infrared (VNIR) to short-wave infrared (SWIR) and thermal infrared (TIR) scans, along with corresponding RGB images of the drill cores, were taken using the HyLogger system. These were made available by AuScope and National Virtual Core Library. The availability of both the line-scan hyperspectral data and the RGB images provides the opportunity to create a combined mineral and lithology classification. Two forms of convolution neural network, AlexNet and MobleNet v2, are implemented for mineral and textural classification. A validation accuracy of ~97% was achieved for mineral classification using the VNIR-SWIR and TIR data. Lithology classification by from RGB achieved a maximum validation accuracy of 94% with AlexNet. When texture prediction was performed on drill cores without ground truth, the results demonstrated a sensible and consistent classification. This study demonstrated that both mineralogical and textural information can be obtained automatically from Hylogger scanned drill cores. This provides a fast, consistent way of gaining additional information that can be used in interpreting and understanding a deposit.

Bio:

Shaoqiu Zheng is currently working on the MPhil project in the scope of hyperspectral lithology study under the supervision of Katie Silversides and Mehala Balamurali. Shaoqiu has been conducting the research in collaboration with Rio Tinto Centre for Mine Automation. This project focuses on deep learning approaches to characterise Ernest Henry Cu-Au deposit drill cores in both hyperspectral and lithological aspects using datasets collected by the Hylogger spectral scanner.

Contacts

Sydney Institute for Robotics and Intelligent Systems
info@acfr.usyd.edu.au